Scholarlink

<p>The detection, classification, and identification of floating vessels from space-borne optical images (VDSBOI) play a pivotal role in various maritime domains, ensuring maritime awareness and coastal security. Despite advancements in vessel detection using Synthetic Aperture Radar (SAR) images, their limited sensors, poor resolution, and long revisit times hinder their real-world applicability. With the advent of high-resolution optical satellite images and advancements in deep learning, VDSBOI has gained prominence, despite challenges posed by environmentally induced noise. This study proposes YOLOGen, a lightweight model that leverages Generative Adversarial Networks (GANs) to mitigate cloud effects and improve detection efficiency. YOLOGen employs YOLO algorithms for detection and classification, achieving state-of-the-art accuracy and speed. Using the YOLOv8 model, we obtained mean Average Precision (mAP) values of 0.625 (@IoU=0.50:0.95) and 0.698 (@IoU=0.5) across 50 vessel classes, demonstrating its potential for real-world maritime surveillance systems. The proposed framework operates at a total inference latency of less than 5.46 ms, achieving a throughput of approximately 183 FPS on an NVIDIA Tesla V100 GPU.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced Detection and Classification of Floating Vessels in Optical Satellite Images via YOLOGen Framework

  • 期刊论文

  • 发表日期：

    2026年5月14日

查看全文
  • Retheesh Vv,
  • Navneet Kaur,
  • Shamla Beevi,
  • Saidalavi Kalady

摘要

The detection, classification, and identification of floating vessels from space-borne optical images (VDSBOI) play a pivotal role in various maritime domains, ensuring maritime awareness and coastal security. Despite advancements in vessel detection using Synthetic Aperture Radar (SAR) images, their limited sensors, poor resolution, and long revisit times hinder their real-world applicability. With the advent of high-resolution optical satellite images and advancements in deep learning, VDSBOI has gained prominence, despite challenges posed by environmentally induced noise. This study proposes YOLOGen, a lightweight model that leverages Generative Adversarial Networks (GANs) to mitigate cloud effects and improve detection efficiency. YOLOGen employs YOLO algorithms for detection and classification, achieving state-of-the-art accuracy and speed. Using the YOLOv8 model, we obtained mean Average Precision (mAP) values of 0.625 (@IoU=0.50:0.95) and 0.698 (@IoU=0.5) across 50 vessel classes, demonstrating its potential for real-world maritime surveillance systems. The proposed framework operates at a total inference latency of less than 5.46 ms, achieving a throughput of approximately 183 FPS on an NVIDIA Tesla V100 GPU.